Fruit harvesting drone based on YOLOv4-tiny algorithm

This study conducted experiments and tests to validate the performance of a mango picking drone. The results demonstrated that the utilization of the YOLOv4-tiny object detection model accurately detects and locates ripe fruits within orchards. The accelerated inference computation provided by TensorRT enhances the real-time capability, enabling unmanned aerial vehicles (UAV) to make prompt harvesting decisions. The three-point localization method calculates distances to ensure precise measurements between the UAV and the target fruits, facilitating more precise and efficient harvesting actions. Additionally, the implementation of Kalman filters in distance detection further improves the stability and accuracy of the system, effectively mitigating the impact of sensor noise and uncertainty.